Importing Ollama models is broken

[dadmobile]

Reported by discord user Rich:

I had imported a downloaded model via Ollama. When I clicked on the "Import Local Model" there were models available to import. After I had done so, I selected one of them and then when I tried to run it, I got the error in the second image. I deleted it and tried importing again, same error.

[dadmobile]

Confirmed that this doesn't work for me either, but in my case the error message included the path to my model:

[dadmobile]

The problem is that there's a colon in the directory name (it should be a dash).

[dadmobile]

Fixed the first part. There seems to be 2 remaining issues currently breaking this:

1. We changed the model ID to something prettier than the ugly SHA, but there is code somewhere that requires that GGUF model IDs and the actual filename are the same. This error manifests as:
   

2. If you get by issue number 1, I am finding that all of my ollama GGUF models are throwing the following in the API:

asyncio.exceptions.CancelledError: Cancelled by cancel scope 31dc562d0

But you can still export these models to llamafile and they work! Probably need to try running these through LLama.cpp and seeing what the underlying issue is.

[delfireinoso]

When I select Import Models, the list includes all versions as one name, but in my ollama installation there is no Qwen2.5 but Qwen2.5:7b and other weights (to test them in different situations). And it's unable to find them.

[dadmobile]

Issue number 2 I mentioned above is fixed by d5bdb9c.

The only remaining issue is the misaligned model names.

I may separate delfireinoso's comment out in to a separate issue. Let me reproduce first to see.